In presidential primaries, a voter may agree with one candidate on the issues, but see another candidate as more ‘electable’ – better positioned to beat the candidate of the other party in the fall campaign. How does that voter decide whom to vote for? Political scientists have found evidence that voters do engage in a judicious weighing of issue agreement and electability. But what affects that weighting?

During this year’s primary campaign in New Hampshire we conducted a survey experiment to find out. Over a four-week period, at dozens of Granite State campaign events for Democratic and Republican Party candidates, we distributed flyers with web links to a short survey on three current political issues. We also asked each respondent to select a ranking on a nine-point scale of whether they based their vote choice mostly on issues or mostly on electability.


Clockwise from top left: Republican presidential candidate Ted Cruz hugs a supporter on January 29, 2016 in Emmetsburg, Iowa (Joshua Lott/The Washington Post). Former Secretary of State Hillary Clinton meets Iowa voters at a rally in Cedar Rapids, Iowa, January 30, 2016. (Melina Mara/The Washington Post). Democratic presidential candidate Bernie Sanders speaks to students at Roosevelt High School January 28, 2016 in Des Moines, Iowa. (Joshua Lott/The Washington Post). Republican presidential candidate Donald Trump at Drake University on January 28, 2016 in Des Moines, Iowa.
(Charles Ommanney/The Washington Post)

Our experiment involved manipulating the presentation of the issues we asked about in the first part of the survey. Half of our respondents received questions about abortion, gun laws, and Muslim immigrants, phrased neutrally. For example: “Do you support or oppose a law that would prohibit most or all abortions?” The other half received questions on the same issues phrased more provocatively: “Do you support a law that would prohibit most or all abortions because abortion is murder or do you oppose that law because abortion is a woman’s fundamental right?”

We expected that respondents who received the more provocative questions would be more likely to view issue agreement as more important than electability in their choice of candidate. We saw this as adhering to previous findings in political science and psychology. Moralized attitudes, for example, seem to lead to less willingness to accept compromising politicians.

Attendees at primary events would be expected to be among the most politically active, most politically interested, and best informed. Therefore, they should be among the least susceptible to the effects of issue framing. If we saw an impact of issue framing in this population, we reasoned, the rest of the population would definitely be receptive to it as well.

What did we find? To sum up:

  • Provocative issue framing makes a significant difference.
  • But the effect is the opposite of what we expected.
  • And the effect disappears in the final days of the campaign.

The effect of provocative issue questions on weight given to electability vs. issue agreement, New Hampshire primary respondents
Data: Middlebury New Hampshire Survey, N=470.
Figure: Bertram Johnson

The technical stuff

The figure above sums up the relationship that we discovered. Here we plot respondents’ answer to the question (on a scale of 0 through 9), “When deciding which candidate to vote for, is it more important that the candidate agree with you on the issues, or is it more important that the candidate has the best chance of being elected (electability)?” Blue dots represent respondents to the control survey, red dots represent respondents to the survey that asked about issues in a more provocative manner. For ease of interpretation, we also plot fractional polynomial fit lines for the two different categories of respondents.

Overall, for attendees at New Hampshire campaign events, anyway, issues are more important than electability. Both groups of respondents are closer, on average, to the ‘issues’ position on our scale than the ‘electability’ position.

The different issue frames in our prior questions did indeed affect the electability/issue agreement calculus, even for this highly politically attentive population – but respondents who received the provocative issue frames were more likely to value electability. This relationship disappears, however, as the election grows closer.

A natural cut point between early in the New Hampshire campaign and late in the campaign occurs in the few days surrounding the Iowa Caucuses, when candidates largely left New Hampshire. This gap is evident in the figure above. We therefore divide our analysis below into pre-Iowa and post-Iowa periods.

As the figure below shows, those who received provocative issue questions before the Iowa Caucuses were more likely to value electability by an average of 0.63 units on a nine-point scale. This difference represents slightly more than one third of a standard deviation in this scale, and is very unlikely to have occurred by chance (p=0.02).


Boxplot for electability scale – neutral vs. provocative issue questions, New Hampshire respondents (pre-Iowa caucus data only)
Data: Middlebury New Hampshire Survey, N=156.
Figure: Bertram Johnson

The finding is all the more striking because we see no difference between the two groups of survey-takers in a separate question about their propensity to volunteer for candidates – something of a surprise, since intensity of feeling about issues is often linked to political participation.

The difference between those exposed to neutral and provocative questions is smaller for those who took our surveys later in the campaign, and diminishes to nothing in the days after the Iowa caucuses took place on Feb. 1. As the figure below shows, this change can be entirely attributed to all respondents moving up on the electability scale to the area where only those who received provocative issue questions had been before.


Boxplot for electability scale – neutral vs. provocative issue questions, New Hampshire respondents (post-Iowa caucus data only)
Data: Middlebury New Hampshire Survey, N=314.
Figure: Bertram Johnson

What does it all mean?

How should we interpret these findings? Our next steps will be to test several theories on our dataset and other datasets to find out.

One possibility is that early New Hampshire campaign events attract a different population than do later events. Our data show that later respondents are slightly older (44 vs. 40, on average), and slightly less likely to have a college education than early respondents. There is no detectable gender or income difference.

We suspect that demographic differences between early and late respondents cannot explain our findings. So here’s an alternative explanation. Perhaps raising the provocative issue increases the stakes of the election in our respondents’ minds. They fear the consequences of a victory for a candidate of the opposite party. Therefore, whether their party’s nominee is electable becomes a greater concern.

But why does the framing effect vanish in the final days of the campaign? Perhaps because near the finish line, the campaigns are all working overtime to do what we did in our survey: provocatively frame the issues and raise the stakes of the election. After Iowa, almost all the candidates were focusing their personal and advertising attention intensely on New Hampshire. The ‘treatment effect’ diminishes because everyone becomes part of the treatment group.

This new evidence may shed light on last-minute shifts in public opinion polls about elections. It could also provide one reason why even the most ‘electable’ candidates might see a reason to use extremist rhetoric. This may provide one more piece to the puzzle of why American politicians have become so polarizing in recent decades.

Bertram Johnson is an associate professor of political science at Middlebury College in Middlebury, Vermont. Margot Graham, Nora Lenhard, Hazel Millard, and Andrew Plotch are students at Middlebury College.